ES2829269T3 - Watermark Embedding and Removal Procedure to Protect Documents - Google Patents

Abstract

A procedure (11) to mark documents with a watermark, receiving an original document (10) that includes a text and a message (30) to obtain a document with a watermark (20) by embedding a watermark that encodes the message received (30), the procedure comprising: - identifying and locating the spaces in the original document (10) by means of a location analysis module (110), distinguishing the location analysis (110) between intra-word spaces and inter-word spaces words, and in which the location analysis (110) comprises: - checking (401) if the received original document (10) is a bitmap formatted document and if the original document (10) is in a non-formatted format. bitmap, converting (402) the original document (10) into a bitmap formatted document; - applying image-based segmentation (403) to the bitmap-formatted document to obtain a set of lengths and a set of locations of the spaces in the bitmap-formatted document; - classify (405) the spaces identified and located in intra and inter-word spaces to obtain sets (406) of lengths of the intra and inter-word spaces, SS and SB respectively, and sets of locations of the intra and inter-word spaces. words, LS and LB, from the original document (10). - minimize the probability of error corresponding to an interference between intra-word spaces and inter-word spaces in the watermarked document (20) by means of an optimization module (111), in which the optimization module ( 111) determines a distance between the lengths of the intra- and inter-word spaces, and in which the optimization module (111) minimizes the probability of error by: - obtaining a modified set of intra-word spaces, S * S = {S * Si} where i denotes a line number of the original document (10) and S * Si is equal to: μ (i) + ε if SSi> μ (i) + ε, and SSI the opposite; where μ (i) is an average value of the SSI intra-word spaces on the ith line and ε is a guest rejection parameter that is an adjustable parameter to increase the distance between the lengths of the intra and inter-word spaces, - obtain a modified set of interword spaces, S * B = {S * Bi} where i denotes a line number of the original document (10), where S * Bi = {S * Bi (k)}, where S * Bi (k) the modified length of the k-th interword space of the i-th line and SBi (k) the original length of the k-th inter-word space of the i-th line of the original document (10), S * S = {S * Si (k)}, where S * Si (k) is the modified length of the k-th intra-word space of the i-th line and SSi (k) the original length of the k-th space intra-words of the i-th line of the original document (10), and S * Bi (k) = SBi (k) + (ΣkSSi (k) - ΣkS * Si (k)) / Nb (i) where Nb ( i) the number of interword spaces in the i-th line of the original document (10). - encoding (113) the message (30) in keywords and encoding (114) the keywords in the watermark; - embed the watermark in the original document (110) to generate (115) the document with the watermark (20) modifying the intra and inter-word spaces of the original document (10) according to the encoding (113) of the received message (30), in which generating (115) the document with a watermark (20) comprises: - selecting a subset of the set of inter-word spaces, S * B, mapped to the set of inter-word spaces of the t-th line, St B, through a secret key; - select a keyword wt from the encoded message (30) - embed the selected keyword wt in the selected subset of interword spaces by calculating a modified interword space SBW from the watermarked document (20) using the expression St BW ( k) = St B (k) + wt (k) -ct (k) -St B (k) = St B (k) (1 + wt (k) -ct (k)) where wt (k) is the selected keyword, for the k-th interword space on the th line; St B (k) is the k-th interword space on the t-th line of the SB interword space of the original document (10), St BW (k) is the k-th interword space on the line t-th of the SBW interword space of the watermarked document (20) and ct (k) is a weighting factor; - modify the intra and inter-word spaces in the original document (10) by substituting the intra-word space, SS, and the inter-word space, SB, respectively by the modified set of intra-word spaces, S * S, and the modified interword space, SBW, in the watermarked document (20).

Description

DESCRIPTION

Watermark Embedding and Removal Procedure to Protect Documents

Field of the invention

The present invention has its application in the telecommunications sector, in the field of digital information security and digital content processing, and is related to digital watermarking technology.

More specifically, the present invention relates to a method for embedding a digital watermark in a document, which can be in a digital or digitized format, including text; and a method for extracting the watermark from the document embedded with the watermark even in case of distortion in the document.

Background of the invention

Including hidden information in a text file is a desired feature in many fields and for many different applications. On the one hand, hidden information embedded in a text document can be decoded by machines to allow automatic classification of documents. At the same time, that information can allow automatic validation of the authenticity, integrity and ownership of a document. In addition, this information -included in confidential documents- can allow its monitoring, deter the leakage of information or help to identify the person responsible in a case of leakage.

The information that is included in documents for this purpose is usually intended to be imperceptible to humans, while it is appreciable to computer systems, which can read and decode it. This inclusion of information is commonly called a watermark. A watermark is an alteration of a document that can include an image or identification pattern, such as the spacing between characters or the deformation of the same in the case of text, or the displacement of the pixels in frequency or space in the case of the images. The watermark can represent a code or a unique sign or signature. The watermark can be reproduced in different parts / sections of the document itself, allowing it to also be extracted from a part of the original document (in the case of damaged / dirty / wrinkled documents). Lastly, the watermark cannot be appreciated by human inspection, while it can be identified in a digital inspection, recovering the represented code.

There are many different procedures for including watermarks in multimedia data, namely audio, video, and images, for which there is extensive scientific literature that describes embedding and extraction procedures. However, the inclusion and extraction of watermarks from multimedia data often involves techniques not applicable to text documents.

Different procedures are used to include information in text documents, often involving adding or changing the appearance of a document. The easiest procedures include visible marks on the document, such as barcodes. However, adding visible marks on the document may not be acceptable for various applications, either for aesthetic reasons or because the watermark must go unnoticed by the user. For this reason, imperceptible watermarks are of great interest for text documents, where there are usually fewer areas in which to hide a mark than in multimedia data.

With regard to imperceptible watermarks in text documents, different techniques are available which are mainly based on one of the following three encoding methodologies: i) Special elements may be involved to embed the watermark, which they are not part of the standard printing process (for example, inks or special papers), ii) the information can be encoded in the text part of the document, using semantic or syntactic rules, that is, altering the content of the text using synonyms or different sets of uppercase and lowercase letters, etc., or iii) information can be encoded in modifying the appearance of the document, including color variations, transpositions of parts of the content, etc.

Some of the techniques available to integrate watermarks in documents are valid for paper documents, others for digital documents (a digital document is an original document created in a digital format, for example, usually PDF), some for Both formats, but all of them have limitations, such as dependence on the printing and scanning process, or the lack of resistance to printing, distortion or the limited amount of information that can be incorporated into the document.

US2005 / 039021 discloses a text watermarking method that incorporates an ancillary message into an original electronic text document to form a watermarked text document. The method applies a broadcast function on message symbols to broadcast the symbols on a modulated bearer and bearer map elements to match interword spaces in the electronic text document. An embedding function then modifies interword spaces to hide the modulated carrier in the watermarked text document. A supported decoder extracts the stub from a watermarked electronic or printed text document. The decoder estimates elements of a modulated carrier signal embedded in the interword spaces and applies a broadcast function to the estimated modulated carrier signal to extract the symbols from the message.

Document US2011 / 016388 discloses a method that incorporates covert data into a text document using space encoding that changes the interword and / or intercharacter spacing in a text line to a particular format. The procedure determines a horizontal space between characters by altering them, so that the altered space represents the embedded covert data. The formatted document is based on altered space so that the data is essentially visually hidden in the text document.

Therefore, it is highly desirable to develop a procedure to embed watermarks in paper text documents and digital documents that are resistant to document distortion, without requiring the original document for watermark extraction and without require any special equipment to print or scan the document.

Summary of the invention

The present invention solves the aforementioned problems and overcomes the working limitations previously explained by the state of the art by providing a method for embedding / extracting digital watermarks to protect documents. More particularly, a method is provided for embedding a digital watermark that encodes a message in an original document. The original document can be a digital or digitized document (a digitized document is a scan / image of a digital document previously printed on paper, or the conversion to a different digital format of a digital document), including text documents, both in vector format as in pixel mapping objects. The reverse procedure is also provided, that is, the procedure to extract the message from the watermark embedded in a document, without requiring the original document.

One aspect of the present invention refers to a method for embedding watermarks in documents, which receives as input an original document and a message to obtain as output a document with a watermark that has an embedded watermark, which encodes the received message. The watermarking procedure comprises the following steps:

- identify and locate the spaces in the original document by means of a location analysis module, distinguishing the location analysis between intra and inter-word spaces;

- minimize the probability of interference error between intra-word spaces and inter-word spaces in the watermarked document by means of an optimization module;

- encode the message in keywords and encode the keywords in the watermark;

- embed the watermark to generate the document with a watermark modifying the intra- and inter-word spaces of the original document.

Another aspect of the present invention refers to a procedure for extracting watermarks from documents, the procedure that receives as input a received document resulting from eventual deformations of a watermarked document, and the procedure that comprises the following steps:

- identifying and locating the spaces in the received document by means of a location analysis module that distinguishes between intra and inter-word spaces and obtains a vector S ' ^bw corresponding to the inter-word spaces of the watermarked document;

- mapping the interword spaces of the vector S ' ^bw to symbols according to an alphabet of keywords w = {wi}, i = 1, 2, ... Q; Q is the number of symbols in the message encoded by the watermark of the watermarked document and the symbols are uniquely mapped to the keywords w;

- unpacking the mapped symbols to obtain sync blocks and payload symbol blocks;

- decode the channel of payload blocks to extract the message that was encoded by the watermark, the watermark embedded in the document with the watermark.

Yet another aspect of the present invention relates to a computer program and a computer-readable non-transitory medium, which contains instructions or computer code (stored in the computer-readable non-transitory medium) to make the processing means (of a processor computer) perform the steps of the watermark embedding and / or removal procedures described above.

The present invention has several advantages over the state of the art, which can be summarized as follows: - Blind embedded watermark retrieval, that is, embedded message retrieval does not require the presence of the original message.

- Resilience to document distortion such as partial interruption, print / scan distortions, color shift in print / scan process, etc.

- Independence of the watermark procedure with respect to the document format (digital, digitized or printed).

- Independence of the watermark procedure with respect to printing and scanning technology, not requiring any specific device or technology.

- The resulting solution is invaluable on human inspection, but identifiable on digital inspection. - No text editing required.

These and other advantages will be apparent in light of the detailed description of the invention.

Description of the drawings

In order to help understand the characteristics of the invention, in accordance with a preferred practical embodiment thereof and in order to complement this description, the following figures are attached as an integral part thereof, having an illustrative character and not limiting:

Figure 1 shows a generic block diagram of the watermark embedding procedure.

Figure 2 shows a generic block diagram of the reverse procedure for watermark extraction. Figure 3 shows a detailed block diagram of the watermark embedding procedure, according to the preferred embodiment of the invention.

Figure 4 shows a block diagram of the location analysis module included in the watermark embedding procedure shown in Figure 3 above, according to a possible embodiment of the invention. Figure 5 shows a message packaging for the watermark embedding procedure of Figure 3, according to a possible embodiment of the invention.

Figure 6 shows a detailed block diagram of the procedure for the extraction of watermarks, according to another possible embodiment of the invention.

Preferred embodiment of the invention

Of course, embodiments of the invention can be implemented on a variety of architectural platforms, operating and server systems, devices, systems, or applications. Any particular architectural arrangement or implementation presented herein is provided for the purpose of illustration and understanding only and is not intended to limit aspects of the invention.

Figure 1 shows that the procedure for embedding the watermark (11) proposed here takes as input an original document (10) and a message (30) to be encoded in the watermark. The original document (10) may be a digital document or a digitized document, including, for example, a scanned document from printed paper. The watermark embedding procedure (11) returns as output a new version of the document, the document with the watermark (20), which is not distinguished in human observation from the original document (10) but which includes the watermark that encodes the input message (30).

The watermark extraction procedure (12), shown in figure 2, is the reverse of the aforementioned watermark embedding procedure (11) and takes as input a copy of a received document ( 60), which can be in digital or digitized format, possibly including distortion, and extracts the embedded message (30). The input element in the extraction procedure (12) is not usually the watermarked document (20), since the received document (60) may be different from the one sent due to distortion, noise, etc.

The watermark embedding procedure (11) is divided into more detailed threads in figure 3. The original document (10) is first subjected to a localization analysis (110), which aims to find and analyze the spaces present in the text of the original document (10). Once the spaces have been located and analyzed, the original document (10) is subjected to an optimization module (111) to minimize the probability of error in a future watermark extraction process, taking into account certain distortion limitations (112) . In parallel, the message (30) to be incorporated into the document must be encoded (113) in keywords and the encoded keywords (114) in a watermark. Finally, the watermark is embedded in a document to generate (115) a watermarked document (20).

The location analysis (110) consists of different steps that are shown in more detail in figure 4. First, it is checked (401) whether the original input document (10) is provided in bitmap format or not. . If the original document (10) is provided in a non-bitmap format, it is converted (402) to a bitmap format to have a common reference for the positioning of the space in the document, in pixels, regardless of the original document (10). As a second step, image-based segmentation (403) is performed to find the spaces in the text of the document. Image-based segmentation (403) can be performed using different techniques, including projection and clustering. Projections represent an easier procedure, while grouping or optical character recognition, OCR, can represent valid solutions in case of non-homogeneous text background. Assuming a standard horizontal distribution of the text in a homogeneous background document, horizontal projections of the text are first made to detect the lines of the text.

Binarizing the document in bitmap format and adding the pixels in the horizontal direction, the horizontal projection Ph (r) is obtained as:

Ph (r) = IcI (r, c)

where r is the rth row, c is the cth column, and I (r, c) is the binarized image of the document.

By setting a suitable threshold value, the document in bitmap format can be segmented into lines of text, finding optimal binarization thresholds from the image histogram. Once the projections are calculated, and lines of text are identified, each line of text is segmented. Each line corresponds to its vertical projection Pvi (c):

Pvi (c) = ZrI¡ (r, c)

where, Ii (r, c) is the binarized image of the document cropped on the i-th line. Applying the same technique used to isolate the text line using the horizontal projection Ph (r), the spaces in each line can be identified and measured by applying a suitable threshold to the vertical projection Pvi (c).

The result 404 of segmentation 403 is a set of spatial lengths and a set of their corresponding locations in the document, denoted as set S and set L respectively. The identified spaces are then classified (405) into intra- and interword spaces. As the process is a bitmap document, this spatial classification (405) can be based on OCR techniques - Optical Character Recognition - or on a grouping algorithm by analyzing the histogram of the spatial lengths, S. As a result of this classification (405), the original sets of spatial lengths S and locations L are divided respectively into sets (406), Ss and Sb, and Ls and Lb, the sets being the length and location of the intra and interword spaces, respectively.

As a next step, the output sets (406) of the spatial classification (405) are fed to an optimization process, in order to reduce the error probability of the future watermark extraction process. In particular, in order to minimize the possible interference between inter-word and intra-word spaces in the watermark extraction process (that is, the interpretation of a space of one type by another type), you can increase the distance between the lengths of the two types of spaces. This operation is called "host rejection". This operation introduces a distortion in the document (10), called Dr, which is regulated by a suitable restriction parameter received as input, called Dr_max. In particular, a possible solution to minimize the probability of error by keeping the distortion below Dr_max consists in taking a modified internal word space S * s, for the i-th line, equal to:

j (i) £ if Ss¡> j (i) S

Ss, otherwise

where | j (¡) is the mean value of the length of the intra-word spaces Ss¡ on the i-th line, and £ is the guest rejection parameter, which must be adjusted to obtain Dr = Dr_max.

Dr is a value indicative of the change in the lengths of the intra-word spaces for all lines of the document (10), that is, Dr = £ i | S * s¡ - Ss¡ |

Hence, Dr = £ i (Ss¡ - (j (i) s)) for Ss¡> j (¡) s

Since Ss, and j (,) are known, the guest rejection value £ is extracted and used to determine Dr = Dr_max.

The effect of the host rejection request mentioned above is to reduce the largest intraword spaces on each line, resulting in larger intra-word spaces. In general, the goal is to keep the total length of each line unchanged. As such, the following condition must be met:

IkS * s¡ (k) I kS * Bi (k) = IkS s (k) XkSBi (k)

where Ssi (k) is the length of the k-th interword space of the i-th line, S * si (k) the modified length of the same space, and, correspondingly, SBi (k) and S * Bi (k) the original and modified length of the k-th interword space of the i-th line.

Being Nb (i) the number of interword spaces in the i-th line, the above equation can be solved using modified spaces equal to:

S * Bi (k) = SBi (k) dkSsi (k) - IkS * s (k)) / Nb (i)

which results in a homogeneous redistribution of the extra space between the inter-word spaces, which in turn results from the reduction of the intra-word spaces.

The message (30) to be encoded in the watermark is a sequence of bits of length K. The message (30) is encoded by an encoding module (113). As a first step, the encoding module (113) adds channel encoding to the message (30) to make it robust to decoding errors. This is achieved by standard forward error correction codes, resulting in a modified message (m ') of length L> K, shown in Figure 5. In a second stage, the modified message (m' ) is divided as a sequence of symbols (500, 500 ') into payload blocks (510, 510') of length N. In addition, before each payload block (510, 510 '), a block of timing (520, 520 ') of length T to allow the watermark decoder to recover from timing errors resulting from misinterpretations of intra- and inter-word spaces, resulting from document distortions . The resulting sequence (m ") has a length L (1 + T / N).

Once the encoded message has been generated, the document with the watermark (20) (115) is generated using a two-step process:

1. A watermark encoding module encodes each symbol of the encoded message in a sequence of spatial lengths. As a result, the interword spaces in the SB * document are changed to the Sbw watermarked interword spaces.

2. The final document with a watermark (20) is generated by modifying the spaces of the previous document, according to the vectors of intra-word space Ss * and inter-word space Sbw.

The embedding of the watermark is done in packets according to a secret key, resulting in a sequence of keywords {wi, ... wq}, where Wi = [wi (1), ... Wi ( P)], i = 1, 2, ... Q, and Wi (k) = ± 1. The number of keywords Q being equal to the number of different symbols in the encoded and packed message, each symbol can be uniquely assigned to a keyword. Then, for each symbol of the package p (i), the following operations are performed:

1. Select a subset of elements P in Sb * of length N I, mapped to S ^ through a secret key. The subset can include elements belonging to different lines of text.

2. Embed the selected keyword wt into the selected subset using the following formula:

SW (k) = S * B (k) Wt (k) ct (k) S'B (k) = Sfe (k) (1 wt (k) ct (k))

where ct (k) is a weighting factor. Depending on the sign of the selected keyword wt (k), for the k-th interword space on the t-th line, the SW (k) interword space of the watermarked document (20) can be longer or shorter than the original space S ^ (k).

The function of the weighting factor ct (k) is to ensure that the first and last letters of each line remain in the same position (really important, for example, in justified texts and to avoid layout modifications in the resulting document) . The weighting factor ct (k) is necessary when the number of large and short interword spaces in a given line is not equal. In general, this is so when

Ik SW (k) = Ik S'B (k)

or, equivalently,

IkWi (k) ci (k) SBi (k) = 0

where ci (k) and Wi (k) are the weighting factor and the component of the keyword respectively for the k-th interword space on the ith line, according to the cartography defined above.

As such, the above condition can be rewritten as:

Ik3 wi (k) = ¹ ci (k) SBi (k) = Ik3 wi (k) = ^-1 ci (k) SBi (k)

the first term being the sum of the spaces for which the coding elements is 1, weighted with the corresponding weighting factor, while the second term is the sum of the spaces for which the coding elements is -1, weighted with the corresponding weighting factor. This is equivalent to:

ci / c ^.1 - Xk3 wi (k) = -1 SBi (k) / Xk3 wi (k) = 1 SBi (k)

Also, if the interword spaces in each line are uniform, which is the common case, then SBi (k) - Sb and the condition becomes: c ¹ / c ^.1 - N ^-1 / N ¹ , where N ^-1 the number of encoding elements "-1" and N ¹ the number of encoding elements "1", in Wi (k) for the ith line.

The generation of the document with the watermark (20) therefore concludes by modifying the spaces of the original document according to the sets Ss * and Sbw.

The reverse procedure (12) to extract the embedded message from the watermarked document (20) is shown in figure 6. Once a document (60) is received, certain operations must be performed to retrieve the embedded watermark . The received document (60) may have undergone a series of transformations since its creation as a watermarked document (20), including printing and / or scanning, eventually introducing distortions in it, which must be taken into account in the process. decoding. Thus, the first step of the watermark extraction procedure (12) consists of noise reduction (610) and geometric correction (620), for example, rotation, scaling and perspective transformation. General distortions include dirt, blur and non-uniform shading, which can be eliminated by applying binarization procedures and morphological operations on the received document (60). The main objective of the noise reduction process (610) is to clean the spaces between words and letters. On the other hand, the main objective of the geometric correction process (620) is to apply the appropriate rotation to the received document so that the lines are approximately horizontal to allow the detection of lines and spaces. To achieve this objective, standard destabilization procedures can be applied, usually based on the detection of edges or transformations to detect the lines of the document and calculate the angle of rotation to apply. The next step of the watermark extraction procedure (12) consists of a location analysis (630) similar to the location analysis (110) described in Figure 3 for the watermark procedure (11) , in order to detect the location and size of the spaces in the document where the watermark is inserted. The spaces must then be classified into intra- and inter-word spaces. If grouping procedures are used for this classification, it must be taken into account that the interword spaces in the received document (60) have two different lengths, different from those of the original document (10), due to the embedding of the mark of Water. Finally, as intra-word spaces do not contain watermark information, they are not relevant as a result of this classification process, while inter-word spaces are recorded in the corresponding SW vector. As an additional step, the spaces in S 'bw have to be mapped (640) to symbols according to the previously defined alphabet w and as follows:

- As the first step of the symbol mapping (640), the spaces are grouped into subsets of size P. The t-th subset of spaces S W will therefore encode a symbol Wi, with i e {1, ..., Q}. The most probable symbol estimate can be expressed as:

c - arg maxc e { ¹ , ..., q> f (SW | Wc)

where f (SW | Wc) is the probability density function of the SW conditions in Wc. In practice, a statistical estimation is difficult to apply, since it implies knowledge of the statistical distribution of SW and its parameters, which varies greatly depending on the type of document and source, and requires a large P to have a statistically sample. significant. Thus, mapping 640 of spaces in symbols is performed based solely on the means of the observed sample. In fact, after embedding the keyword Wc in SB, the sample mean S W has the form | Jk (1 + AWc (k)), where | Jk is the mean of S W). Therefore, the spaces of the watermarks are distributed around the centers of Q, so that the embedded symbol can be estimated by vector quantization techniques, based on various existing procedures for such operation, including the scalar way. Note that if a grouping procedure has been used in the location analysis (630), that classification can naturally be provided as a secondary result of that process.

- As a next step, the mapping between SBwi (k) and Wi (k) is carried out, assigning Wi (k) the values of 1 or -1 depending on the corresponding spatial classification as large or short, respectively. The estimates of Wi (k) are ordered according to the subset defined in the embedding phase and the resulting sequences are compared with the alphabet keywords. If Wc is the most similar keyword, the symbol c e {1, ..., Q} is selected.

The estimated symbols resulting from the mapping (640) are finally input to an unpacking step (650), which searches for the sync symbols and extracts the symbols from the payload. Finally, channel decoding (650) is applied to payload symbols to extract the original message (30) that was embedded in the received watermarked document (60).

Note that in this text, the term "comprise" and its derivations (such as "understand", etc.) should not be understood in an exclusive sense, that is, these terms should not be interpreted as excluding the possibility that what is described and defined may include other elements, steps, etc.

Claims (10)

A procedure (11) to mark documents with a watermark, receiving an original document (10) that includes a text and a message (30) to obtain a document with a watermark (20) by embedding a watermark that encodes the message received (30), comprising the procedure: - identify and locate the spaces in the original document (10) by means of a location analysis module (110), distinguishing the location analysis (110) between intra-word spaces and inter-word spaces, and in which the analysis of location (110) comprises: - check (401) if the original document received (10) is a document with bitmap format and if the original document (10) is in a non-bitmap format, convert (402) the original document (10) in a bitmap formatted document; - applying image-based segmentation (403) to the bitmap-formatted document to obtain a set of lengths and a set of locations of the spaces in the bitmap-formatted document; - classify (405) the spaces identified and located in intra and inter-word spaces to obtain sets (406) of lengths of the intra and inter-word spaces, Ss and Sb respectively, and sets of locations of the intra and inter-word spaces. words, Ls and Lb, from the original document (10). - minimize the probability of error corresponding to an interference between intra-word spaces and inter-word spaces in the watermarked document (20) by means of an optimization module (111), in which the optimization module ( 111) determines a distance between the lengths of the intra- and inter-word spaces, and in which the optimization module (111) minimizes the probability of error by: - obtaining a modified set of intra-word spaces, S * s = {S * yes} where i denotes a line number of the original document (10) and S * yes is equal to: M (i) £ if Yes> M (i) £, and if the opposite; where p (¡) is an average value of the intra-word spaces Ssi on the ith line and £ is a guest rejection parameter that is an adjustable parameter to increase the distance between the lengths of the intra- and inter-word spaces, - obtain a modified set of interword spaces, S * B = {S * b¡} where i denotes a line number of the original document (10), where S * b¡ = {S * b¡ (k)}, where S * Bi (k) is the modified length of the k-th interword space of the i-th line and SBi (k) the original length of the k- th interword space of the i-th line of the original document (10), S * s = {S * s¡ (k)}, where S * si (k) is the modified length of the k-th intra-word space of the i-th line and Ss¡ (k) the original length of the k- th intra-word space of the i-th line of the original document (10), and S * B¡ (k) = SBi (k) dkSsi (k) - IkS * s¡ (k)) / Nb (i) where Nb (i) is the number of interword spaces in the i-th line of the original document (10). encoding (113) the message (30) in keywords and encoding (114) the keywords in the watermark; embed the watermark in the original document (110) to generate (115) the document with the watermark (20) modifying the intra and inter-word spaces of the original document (10) according to the coding (113) of the received message (30), in which generating (115) the watermarked document (20) comprises: selecting a subset of the set of interword spaces, S * B, mapped to the set of interword spaces of the t-th line, StB, through a secret key; select a wt keyword from the encrypted message (30) embed the selected keyword wt in the selected subset of interword spaces by calculating a modified interword space Sbw from the watermarked document (20) using the expression SW (k) = Styk) wt (k) -ct (k) -S'B (k) = Styk) (1 wt (k) ct (k)) where wt (k) is the selected keyword, for the k-th interword space on the th line; StB (k) is the k-th interword space on line t-th of the interword space S ^b of the original document (10), SW (k) is the k-th interword space on line t -th of the interword space S ^bw of the watermarked document (20) and ct (k) is a weighting factor; modify the intra- and inter-word spaces in the original document (10) by replacing the intra-word space, S ^s , and the inter-word space, S ^b , respectively by the modified set of intra-word spaces, S * s, and the modified ^interword space, S bw, in the watermarked document (20). The method of claim 1, wherein image-based segmentation (403) applies grouping or optical character recognition if the received original document (10) has a non-homogeneous text background. The method of claim 1, wherein image-based segmentation (403) applies projections if the received original document (10) has a homogeneous text background. The method according to any preceding claim, wherein the encoding (113) of the message (30) of length K comprises adding a channel encoding to the message (30) to obtain a modified message (m ') of length L > K, the division of the modified message (m ') into payload blocks (510, 510') of length N and the addition of a sync block (520, 520 ') of length T before each payload block (510, 510 ') respectively to obtain a sequence of symbols (m ") of length L (1 + T / N). The method according to any preceding claim, wherein the original document (10) of a digital document and a digitized document are selected. A method (12) for receiving a document (60) of the method of claim 1 and extracting watermarks from the received document (60) resulting from a watermarked document (20) having an embedded watermark that encodes a message (30) of a number Q of symbols, the procedure comprising: - identify and locate the spaces in the received document (60) by means of a location analysis module (630), distinguishing the location analysis (630) between intra- and inter-word spaces and obtaining a vector S ' ^bw corresponding to the spaces inter-words of the watermarked document (20); - mapping (640) the interword spaces of the vector S ' ^bw to symbols according to an alphabet of keywords w = {wi}, i = 1, 2, ... Q; where Q is the number of symbols in the message (30) and the symbols being uniquely mapped to the keywords w; - unpacking (650) the mapped symbols to obtain sync blocks and payload symbol blocks; - decoding the channel (660) of payload blocks to extract the message (30). The method according to claim 6, wherein the received document (60) is a distorted copy of the watermarked document (20). The method according to any of claims 6-7, further comprising applying noise reduction (610) and geometric correction (620) to the received document (60) prior to location analysis (630). 9. A non-transient computer-readable medium containing computer instructions stored therein for causing a computer processor to perform the method according to any of claims 1-5. 10. A non-transient, computer-readable medium that contains computer instructions stored on it to cause a computer processor to perform the procedure in accordance with any of statements 6-8.